If you are looking to extract more information from the FFT graph you have plotted currently there is not a lot to work with. However, if I were you I'd ask these questions:

• Are these two peaks clear natural frequencies or is there poor resolution?
• If they are clear, what are their amplitudes? All natural frequencies are not necessarily equal the amplitude may shift depending on the stiffness/damping or mass of the specimen's local physical properties.
• Based on the differences in amplitude of the differently placed accelerometers, you may be able to reconstruct a rough simulation of the specimen movement throughout the sweep...? A good visual aid.
• Have you considered extending your analysis to a frequency-time analysis? Or changing your input from a 1G sine sweep to another periodic input pattern?

These maybe very simple questions you have already considered, but I just thought I'd mention it. Not aware of the objectives of your research, so it's difficult to tailor a response.

Possible explanations will depend on the type of signal, sound presure level or admittance directly on the object. The signal at hand may also play a role. For impacts, there may be dips in the response due to the final duration of the impact, a comb filter is formed. A comb filter may also appear due to signal windowing and possible zero padding.

A split resonance in a sound level plot may be the response of two or seveal modes, reflections from near surfaces and thus comb filtering, two close resonances in the measured system, resonances with same reseonace frequency, but different mode shape, a small passive absorber, a version of the former.

A split top ca be a sign of coupled modes, also a differnt way of describing the former phenomena.

Two possible peaks also shows harmonics of the fundamental frequency I.e multiples of the fundamental frequency